FAQ & Critical Review

The black-hat perspective. A specification is only as strong as its ability to withstand scrutiny. Here we address the toughest questions head-on.

Isn't this just more YAML complexity?

FLUID eliminates complexity by unifying scattered configurations. Instead of separate dbt models, Airflow DAGs, data-quality scripts, and access policies, you get one declarative file. Fewer moving parts means less complexity, not more.

Does FLUID replace my existing tools?

No. FLUID makes your tools work better together. dbt, Airflow, Snowflake, and others become "FLUID-aware" by reading the .fluid.yml specification to auto-configure themselves. FLUID is the shared language, not a replacement platform.

How do I start using FLUID today?

Start small:

  1. Pick one critical data pipeline.
  2. Write a .fluid.yml file describing it — see the Quickstart and FLUID by Example.
  3. Use FLUID-compliant tools, or build adapters for your existing stack.
  4. Gradually expand to more data products.

What about complex transformations and custom logic?

FLUID supports multiple build patterns:

  • hybrid-reference — for dbt-style transformations.
  • embedded-logic — for custom SQL / Python code.
  • multi-stage — for complex multi-step orchestration.
  • acquisition — for source-aligned ingestion from external systems.

The lineage block maintains full traceability even with custom code.

How does this help with AI agents and the "agentic era"?

AI agents need contracts, not chaos. FLUID provides:

  • Discoverable data — agents can find the right data products.
  • Trustworthy contracts — schema, quality, and freshness guarantees.
  • Secure access — policy-driven permissions for autonomous systems.
  • Rich context — business semantics and lineage for better decisions.
  • AI governance — model whitelisting, usage quotas, and audit trails via agentPolicy, enforced at runtime at the MCP gateway.
  • Sovereignty controls — automated jurisdictional compliance.
  • Fine-grained permissions — root-level access policies with automated IAM.

Want to go deeper on where FLUID is honest about its gaps? See the Agentic-Native Layer, which tests each spec against its own canonical example.